Electromagnetic Interference (EMI) in Power Supplies Alfred Hesener

Abstract -- Increasing power density, faster switching and higher been widely accepted. The sensitivity to load and line currents forces designers to spend more time both considering the effects of electromagnetic interference (EMI) and debugging a design regulations can limit their usage and parameter that has EMI problems but is otherwise complete. This paper explains variations of passive components can make series the different types of EMI and their coupling mechanisms and the existing EMI regulations. The most frequent noise sources, production difficult and expensive. Further, for some transmission paths and receiver sensitivity are examined. Based on stages of the power supply (e.g. secondary side post- real designs and measurements, specific procedures are recommended for use throughout the design cycle, to make the power regulation) a resonant version does not really exist. It is supply work reliably and pass EMI testing. only with today’s modern control ICs that quasi-resonant power supplies show their potential while maintaining I. INTRODUCTION good EMI performance. So it is not surprising that more In power supplies, the two prominent types of EMI and more designs are using this topology. are conducted EMI and radiated EMI. Comprehensive Given these new developments, it is clear that EMI regulations provide limitations to radiated and conducted performance can no longer be considered only after the EMI generated when the power supply is connected to main power supply design is finished. It needs to be the mains. designed into the power supply right from the start at Comparing the modern power switches used in power specification level, just like reliability and safety, supplies with those from older generations, the new influencing topology and component selection. switches have significantly reduced switching times, The goal is to meet EMI regulations while not leading to faster and faster rise and fall times for the disturbing other applications nearby. The power voltage and current waveforms. These fast edges supply should also be self-compliant and tolerate a produce significant energy at surprisingly high certain amount of EMI from the outside. This paper frequencies, and are the root cause of all EMI problems will show how to “embed” EMI considerations in switched-mode power supplies. This high frequency throughout the entire design cycle. The intent is to energy causes ringing in all the resonant tanks, small or give the power supply designer a good large, that exist within the power supply. In general, this understanding of the problem and an overview of wringing does not cause problems; however, in some the measures that can be taken while designing and cases, this may stop the power supply from working testing the power supply, to improve time to market properly or passing tests. and to come up with a robust design. It is not a Faster switching also means that losses can be reduced, comprehensive overview on the topic, as a large improving the efficiency of the power supply. But faster amount of good literature exists already.[1]-[4] switching should also enable higher switching frequencies, ultimately leading to smaller passive II. DIFFERENT TYPES OF EMI AND components and better transient behavior – a promise THEIR CHARACTERISTICS that has not been realized. Three things can cause an EMI problem: A signal The main reasons for this are the cost of transformers source creates some kind of noise, there is a transmission for use at these frequencies and the disproportional path for the noise, and/or there is a receiver sensitive complexity of solving high frequency EMI problems. enough to be distorted by the noise, as shown in Figure 1. Resonant and quasi-resonant topologies offer an elegant way out of this dilemma. They have been around for a long time, but due to limitations, they have not

Fairchild Semiconductor Power Seminar 2010-2011 1

Shielding the source using thin conductive layers is most effective. Inductive coupling: This transmission path is quite common in switched-mode power supplies since high- frequency currents in the inductors can cause strong magnetic fields at higher frequencies, where the Fig. 1. EMI sources. coupling factors can be higher. Magnetic shielding is The noise source can be inside or outside the power less effective than electric shielding since the absorption supply. Tackling the noise problem at the source means depth is smaller, requiring thicker materials. Inductive reducing the emission levels — for example, by coupling is best addressed at the source. lowering noise amplitudes. Different coupling Wave coupling: Here, the noise typically has a high mechanisms exist for noise, and many EMI frequency, and is transmitted via an electromagnetic countermeasures focus on these; however, they overlook wave. It does not play a major role in power supplies, what can be done at the emitter or receiver. A receiver since frequencies are not high enough, and can be susceptible to noise injection must exist in the system if damped very effectively with shielding. there is an EMI problem. Here, the obvious solution is This paper will focus on capacitive, resistive, and reducing its sensitivity. inductive coupling; as they are the most important At this point, a fundamental distinction must be made sources of EMI issues in power electronics applications. between the two types of EMI problems: It is generally accepted industry practice to consider - Improving EMI so that the design meets conducted EMI below 30MHz, radiated EMI above regulations and will pass EMI testing (also 30MHz, and in most cases up to 1GHz – exceptions do called EMC or electromagnetic compliance) exist, however. - Improving EMI so that the design works reliably Coupling modes cannot be treated in isolation since in all modes of operation, with good efficiency, ideal elements exist only in simulators, not in real life. and does so without being disturbed by other Parasitic elements are always present. The parasitic (EMC-compliant) equipment nearby capacitors and inductors contribute to the problem, as For the first type, test methods and certified labs exist. parts of tank circuits that will resonate when stimulated For the second type, it is important to take the design by a voltage or current edge. The parasitic tanks help to through all design stages, carefully checking to see if convert one coupling mode into another, and that is why poor EMI design may be the cause of the problem. Here, coupling modes cannot be analyzed and fixed in it is important to consider component variations. Maybe isolation. The third parasitic element, resistance, actually the components in the prototype are such that no helps to ease the problem by damping the resonant problem is visible, but the components used in oscillation. Using the amplitude change from peak to production may cause problems. peak can help to calculate the parasitic resistance, The four coupling mechanisms are: identify it in the circuit, and optimize the circuit Resistive (or galvanic) coupling: The noise signal is accordingly. transferred via electrical connections. This works at all frequencies, and is usually fixed by good layout III. REGULATIONS AND STANDARDS FOR (particularly the ground layout) and filtering with EMI capacitors and inductors or lower signal levels with RC As electrical consumers moved from simple light elements. “Common impedance” coupling can be bulbs to large motors and particularly to switched-mode classified as galvanic coupling. power supplies with rectifiers and capacitors at the Capacitive coupling: Electrical fields are the main inputs, the quality of the grid voltage and service transmission path. Capacitance levels are mostly small worsened. This led to the emergence of worldwide so this affects small signals and/or high frequencies.

Fairchild Semiconductor Power Seminar 2010-2011 2 standards to mitigate these problems. Two measured and appropriate limits defined. Using a considerations of these standards are: defined impedance to measure the harmonic current is - Limit the amount of emission not required today, but is also under discussion. (radiated/conducted) which a given application The standard 61k-3-2 defines the limits for generates applications <16A, and 61k-3-12 for 16A…75A current - Define the minimum immunity levels consumption. The standard 61k-4-7 defines the (radiated/conducted) a given application must measurement and evaluation method. Additional 61k tolerate without malfunction standards define the behavior for voltage and frequency The list of standards is very long. The common theme variations, immunity to conducted and radiated radio is that certain standards define the limit values and their frequency signals, fast voltage transients, surges, voltage measurement methods and conventions, and additional dips / drops, short interruptions and flicker, and documents define the regulations for classes of magnetic fields. applications in more detail. For the standard 61k, the equipment typically is Additionally, standards can be grouped into grouped into different classes, with professional local/regional standards, MIL standards, automotive equipment above 1000W power consumption still being standards, standards for the aircraft industry, for excluded. Different limit levels exist for the different physically large equipment, and for more specialized classes. An overview is given below. equipment (e.g. smart meters). In most cases, a good PFC circuit is sufficient to meet The two most important standards for power supplies the 61k regulations. Circuits that have just an input are EN550xx and EN61000. Applications connected to rectifier and capacitor connected directly to the mains the grid must comply with both. The first covers EMI may not be able to meet the requirements without a PFC. limits for various applications, defining the measurement It is also important to note that 61k requires repeatability methods in more detail for both conducted and radiated of the test results, within 5%. EMI, defining limit values, and mostly considering the high frequency content the application generates. The Class Equipment Power Comment 3-phase equipment, household following list gives an overview: appliances, tools, dimmers for Limit values are A incandescent lamps, audio > 75W defined as absolute equipment, everything not B, C values CISPR11, EN55011 for industrial, medical, scientific or D Limit values are Portable tools, Arc welding applications B > 75W defined as absolute equipment values CISPR13, EN55013 for consumer applications Limit values defined CISPR14, EN55014 for home appliances, power tools, C Lighting > 25W as relative values to first harmonic involving motion control Limit values defined only for 3rd and 5th CISPR15, EN55015 for lighting equipment C Lighting < 25W harmonic, relative to CISPR22, EN55022 for computing applications first harmonic Personal Computer, Monitor, Limit values relative D 75W - 600W Television per input power The standard CISPR16 / EN55016 defines the measurement method for the applications listed IV. MEASUREMENT AND SOURCES OF above and is central to all of them. EMI The standard EN61000 (or “61k”) is the “PFC” Conducted EMI standard and considers the line frequency harmonics a th The impedance of an AC power line varies widely. given application generates, up to the 40 harmonic The impedance at the end of a long cable in some remote frequency or 2kHz and below. Extension of the standard area can be quite different from the impedance of a cable to consider interharmonics up to 9kHz is in discussion. in an industrial park, located right next to a transformer No defined impedance is used with the standard but the station. When a power supply generating noise is harmonic current content generated by the application is connected to this impedance, the noise measurement

Fairchild Semiconductor Power Seminar 2010-2011 3 results will vary widely. For this reason, a standardized impedance is used. The noise current generates a voltage across this standardized impedance, so the results for noise levels can be compared. The standardized setup, as defined in CISPR16 / EN55016 for measuring conducted EMI is shown below:

Fig. 3. Equivalent circuit of a LISN. Differential mode noise is caused by the time-varying

Fig. 2. Connection of a LISN to a power supply under test. current demands of the switching stage, conductively coupled via the bus cap into the lines, and appears as an

out-of-phase voltage at the lines. The measurement On the left side, the block called LISN (“Line points showing where to connect the analyzer are Impedance Stabilizer Network” also known as AMN, for indicated in Figure 3. In reality, the two voltages will be “Artificial Mains Network”) represents the standardized quite different, depending on the impedance of the line impedance. The noise signal is taken from this but can be “decomposed” into their differential and impedance with a high-pass filter, amplified, and common mode equivalents. connected to a spectrum analyzer. This analyzer is used An EMI analyzer is shown in Figure 4. Typical to measure the harmonic energy content of the noise specifications include a bandwidth of at least up to signal across a wide frequency range. 1GHz, and a selectable bandwidth to perform the Fig. 3 contains a line impedance stabilizer network to measurements in line with the standards. A large connect the line, load, and the spectrum analyzer. Note dynamic range is important. If the analyzer starts that the connections are “live.” The LISN is directly clipping high-level peaks, the measurement is unreliable connected to the mains, in contrast to the usual due to noise spill-over. measurement practice where an isolation transformer is used for safety reasons. The ground connections can also carry significant leakage current. It is important to realize this difference, as it has significant safety implications when operating the equipment in test. One important differentiation to make is the one between common and differential mode noise. Common mode noise is caused by a capacitive coupling of the switching stage into ground, and appears with the same Fig. 4. Example Spectrum Analyzer used for EMI testing. phase and amplitude at both lines. Two measurement methods exist, called “Average” and “Quasi-Peak”, and they have different limit values as shown in Figure 5 below (QP = upper line in red, AV = lower line in blue):

Fairchild Semiconductor Power Seminar 2010-2011 4

definitions are shown in Figure 6 and defined in standard CISPR-16.

Fig. 5. EMI Frequency Spectrum showing average (AV) and quasi-peak (QP) plots. The frequency range on the x axis is from 9kHz to 30MHz. The noise level on the y axis is scaled from 0 to Fig. 6. Bandwidth definitions for EMI measurement. 100dBµV, with VdBµV = 20log (Vrms/106). The straight red line represents the limit values for the quasi-peak In switched-mode power applications, conducted EMI measurement, and the blue line the limit values for the is primarily caused by fast voltage changes. average measurements (both correspond to the standard Unfortunately, this is how all switched-mode power EN55011/22-ClassB). supplies work. The fast edges contain many harmonics This power supply clearly exceeds the limits at around and these harmonics are coupled into both inputs and 18MHz. Note that the limits are independent of the rated outputs with different damping. The rise and fall times output power of the power supply, explaining why more also influence the high frequency content of the noise. effort needs to be made to meet the limits for larger Reducing the switching speed improves EMI behavior power supplies. The table below shows the limits but reduces efficiency, especially in hard-switched according to EN55022, class B: topologies. In terms of the emitter-coupling-receiver model discussed earlier, nothing can be done at the Limit Frequency Limit (V) Comment (dbµV) receiver since this is the defined impedance of the LISN. 9kHz ... 50kHz 110 316mV Quasi-peak However the switching and coupling can be influenced.

50kHz…150kHz 90 … 80 32mV … 10mV Quasi-peak First, the noise signal coming from the switching Quasi-peak; linearly action has to pass through the DC blocking capacitor. 66 ... 56 2mV ... 0.63mV falling with log This capacitor is a non-ideal element – if it were ideal, it (frequency) 150kHz ... 500kHz Average; linearly would have a very low impedance, approaching zero at 56 ... 46 0.63mV ... 0.2mV falling with log (frequency) very high frequencies and effectively eliminating all of 56 630µV Quasi-peak the noise. Its parasitic characteristics contribute to 0.5MHz ... 5MHz 46 200µV Average reduced damping at higher frequencies, being the main source of conducted EMI in a switched-mode power 60 1mV Quasi-peak 5MHz ... 30MHz supply. 50 316µV Average A capacitor like the (typically electrolytic) bulk capacitor in the SMPS has a parasitic resistance (ESR) Note also that there is an increase in the limits at and a parasitic inductance (ESL). These are typically 150kHz, accompanied by a change in measurement modeled in series with the main capacitor, as shown in bandwidth from 200Hz to 9kHz – this is the main reason Figure 7 (the leakage resistor is of lesser importance here). (albeit not the only one) why switched-mode power supplies usually operate at main switching frequencies below 150kHz. In fact, a switching frequency of less than 50kHz can be desirable, since this would place the second and third harmonic in the frequency space where higher levels are tolerated. The measurement bandwidth Fig. 7. Equivalent circuit model of an electrolytic capacitor.

Fairchild Semiconductor Power Seminar 2010-2011 5

Typical values for an industrial grade 100µF/450V increasing the order of the filter, and works well in most capacitor, like the B43601 from EPCOS, are ESRmax = SMPSs where a large bus cap is connected to the output 1900mΩ and ESL = 20nH. At lower frequencies, of the filter. The T filter has an inductive input and especially the main switching frequency, a current is output, where the input makes life easier for the over- being forced in and out of the capacitor, causing a voltage protection at the input (the voltage can rise faster voltage drop across the ESR. This pulse current and ESR compared to the pi filter where the input capacitor would value can be used to assess the voltage at these have to be charged first). For “difficult” lines, this may frequencies, and comparing it with the allowed levels be a better choice. The output, however, is facing the bus under EN550xx, yields the required attenuation the EMI cap and may starve the SMPS if not designed properly. filter must provide. At higher frequencies, the ESL In many cases, the L filter provides a good compromise impedance is higher than the ESR, and the capacitor but offers only 12dB loss per octave since it has only behaves like an inductor. two elements (a double-L may be required in some For conducted EMI, the basic filter topologies are the cases). Here, the bus cap cannot replace the output cap pi filter, the T filter, and the L filter. The circuits shown of the filter, since its SRF is too low. in Figure 8 are for both balanced and unbalanced versions. The left side is always the line side; the right Schematic Loss per octave pi filter 18 dB / oct side is the load side. Depending on the required attenuation, the filter needs to have one, two, or even three stages (it is rare to see more than two stages). It is important to consider the power and signal flow when laying out the filter. The best form factor is usually achieved with a balanced T filter 18 dB / oct implementation: making the filter long and thin, reducing the coupling capacitance, and increasing the impedance between input and output. The components

should be large enough to cope with the required peak currents and provide sufficient damping (with a margin). L filter 12 dB / oct They should not be too large, since all capacitors and inductors have a self-resonant frequency (“SRF”), which depends on the parasitic inductance and capacitance, and this frequency will be lower with larger components.

Above the SRF, the attenuation function is basically gone. This explains why sometimes using two smaller Fig. 8. Types of input filter. components instead of one large may be better. These filter topologies address differential mode noise It is important to note that the EMI filter will work but not common mode noise. For this, an additional most of the time in an unmatched setup, with the line element needs to be introduced to increase damping on and load impedances different from the design the lines and provide a return path for the noise. The impedance of the filter, which is therefore reflecting common mode noise is primarily caused by capacitive most of the energy. However, the energy must be coupling of the switching stage into the ground line, and absorbed somewhere, underlining the need for lossy the current loop for this noise is then completed via the components in the circuit. As a general rule, it is better LISN and the differential mode EMI filter. The coupling to offer a dissipation path for the unwanted energy, capacitance is typically small, so for the filter to be rather than letting it find its own path. effective, large inductance values are required. The line- A pi filter is used in most applications. The pi filter to-line caps in the differential mode EMI filter do not has an advantage in coupling with the LISN, effectively help here, and the inductors in this filter are too small to

Fairchild Semiconductor Power Seminar 2010-2011 6 provide useful damping. An effective solution to the common mode noise current; and they must comply with problem is the so-called “common mode choke”, where the safety requirements as the ground connection may two inductors are wound on a common core, and break, and a user might touch the midpoint. These connected so that the two inductors look like a capacitors are also called Y-caps, since in most cases transformer with the winding starts connected in phase two are connected from the two lines to ground. for common mode noise (see Figure 9): The calculation of the filter components requires the following data:

Line frequency fLine

Minimal RMS voltage Vmin

Maximum RMS load current Imax

Lowest switching frequency fswmin Fig. 9. Common mode and differential mode choke.

First, the design impedance is calculated as: This inductor is also called the “zorro” inductor. It is designed to have the required inductance value for (1) common mode (the left inductors) but wound in a way that maximizes leakage inductance, providing the Next, the filter topology is chosen based on the required inductance level for differential mode (the right characteristics described above. The number of stages is inductors) – an elegant and compact solution to the determined later. problem. Finally, the required attenuation at the lowest problem Figure 10 shows a practical EMI filter implementation. frequency fswmin is determined. This is either by The black area in the upper middle part is the input simulation or by measurement of the power supply connector. Next to the input connector is the fuse, and without the EMI filter and using the final layout for the the metal box is the main switch. The first choke is rest of the power supply. below the main switch, followed by the first capacitor The filter cutoff frequency fcut is determined from the (gray square), the second choke and the second capacitor. attenuation needed at the problem frequency (plus a margin of 6 … 10dB), while at the same time providing minimal attenuation at the line frequency. This is an iterative process, ultimately determining how many stages are needed in the filter. The cutoff frequency should be at least 10 times the line frequency. Calculate the L and C values of the filter as follows:

, (2) Fig. 10. Practical EMI filter implementation. The purpose of the dark gray resistor is to limit inrush current. Part of the bridge rectifier is visible. This filter Using these component values, the individual values is a two-stage L filter using two chokes. No capacitor is are calculated and the filter is constructed. For the connected across the lines at the input side. differential mode filters, these values are the same as the It is important to connect a capacitor from the lines to components in the above filter circuits. For the common ground to provide a return path for the common mode mode filter, it is important to note that L is the noise current. These capacitors have three requirements. inductance of the common-mode choke windings in They must be small enough to not cause too high parallel and C is the capacitance from the lines to ground. leakage, tripping the ground fault interrupters; they must be large enough to provide low impedance for the

Fairchild Semiconductor Power Seminar 2010-2011 7

Further optimization can be done. Here are some simpler radiated EMI test can be used to spot potential suggestions that are further discussed in the problems, using a so-called “sniffer” probe.[4] references[1][4]: This probe consists of a small coil which is shielded - Insert series-resonant tanks fine-tuned to the against electrical fields and isolated on the outside for problem frequency. This will work only for fixed- safety reasons. Two examples are shown in Fig. 11: frequency SMPS or fixed-frequency noise peaks and the impact will depend on component variations which can be large with passive components, so bandwidth must be increased at the expense of

damping at the design frequency. Fig. 11. 3D and 1D EMI sniffer probes. - Insert RC shunts across the lines, to attenuate certain The first probe has a linear coil, causing it to be most frequencies and also introduce some damping sensitive across the length of the probe. The second - Add resistive impedance into the filter to make the probe has three coils, which are perpendicular to each filter more lossy. other and electrically connected in series. Its sensitivity - Reduce the quality factor (by optimizing the to H-fields is independent of the field orientation. It inductors) of the stages in the EMI filter to reduce depends on the particular case to determine which probe ringing. The ringing frequencies should be far away is best, but in general, the 3D probe is used to “scan” the from the problem frequencies, the main switching application for locations with high field intensity, and frequency and its third harmonic frequency. the 1D probe is used to get a more precise picture of the Capacitors usually have a much higher Q factor than nature of the problem. inductors, and not much can be done about it. - A differential mode choke can help in difficult situations, but is a more complex component and rarely seen in SMPSs. Their current capabilities generally limit their use to lower powers.

Radiated EMI Measuring radiated EMI is also defined in CISPR16 / EN55016 and requires a probe (antenna) for the magnetic field. This can be as simple as a small coil Fig. 12. Using an EMI sniffer probe. connected to a measurement amplifier and to an The test setup with these probes is shown in Figure 12. oscilloscope or spectrum analyzer. The magnetic probes The probe is connected to a probe amplifier and then to used in EMI labs are calibrated probes, and require a fair an oscilloscope. Both the probe and the amplifier output amount of know-how and additional equipment to make require a 50Ω termination. An additional HV probe is accurate measurements. Performing the measurements connected to the board under test. according to the requirements of the standard is Radiated EMI issues in switched-mode power significantly more complex than performing the supplies are generally related to currents being switched. measurements for conducted EMI, since the field probes This usually happens in sync with some clock frequency. need to be calibrated and a measurement room needs to This represents a very convenient way to get closer to be used which is free from other magnetic fields. the problem, simply by using this clock or switching In switched-mode power supplies, radiated EMI signals to synchronize the oscilloscope, so that the image “hotspots” can serve as an indicator for potential becomes stable. Switching a current can trigger resonant problems; however, when fixed, the power supply tanks in the application, and they then resonate with their usually passes EMI testing without any problems, as characteristic frequency. This frequency is from the magnetic fields usually do not reach very far. Hence, a

Fairchild Semiconductor Power Seminar 2010-2011 8 oscilloscope screen and is used to calculate which of the Reduce the coupling factor M between the magnetic parasitic elements in the circuit might be the source. The loops. At system design and PCB level, the orientation speed with which the ringing decays can give an insight of the current loops should be orthogonal, not parallel. into the Q-factor of that resonant tank. The current loop areas should be made as small as As mentioned, this type of EMI comes from switched possible, even if that means running the return path in currents with high di/dt. The two most prominent parallel or on top / under the current path and increasing sources are loops with fast current rise times and the the resistance (we will see later that this might not be leakage fields of inductors (other examples are given such a bad thing). Increasing the distance between the below). The coupling mechanism works like a emitting current loop and the loop picking up the noise transformer, and is shown in more detail in Figure 13. may help, but requires a redesign of the PCB if the problem is detected when the hardware is built. At this point, magnetic shielding can help, but this includes manufacturing complexity, adding to the system cost. It is important to identify potential current loops with high di/dt in the circuit and take appropriate measures at the design level to reduce the impact. To do this, analyze the current flows at normal behavior of the circuit and check which elements only see current flow in one part of the cycle – these elements are very likely to be in a Fig. 13. di/dt induced EMI. current loop with high di/dt. The current change induces a noise signal voltage: Other more unusual sources of resonant ringing and EMI radiation exist, namely: RM di Diode reverse recovery. Once conduction stops, the Vnoise = × M × (1) RS + RM dt space charge needs to be replenished, and in case of “snappy” diodes, the resulting di/dt can be high. The coupling factor M depends on distance, area and Transformer shield ringing. The shield is orientation of the magnetic loops, and magnetic capacitively coupled to the primary winding that sees absorption between the loops – just like in a transformer. high dv/dt, and has parasitic inductance in the path to In addition, noise voltage depends on the strength of the ground. This tank is excited by primary-side switching current change and on the impedance of the receiver. and can couple radiated EMI into the ground connection. This correlation is straightforward and the necessary Ringing between parallel caps. Small differences countermeasures can easily be derived. (e.g. in the ESR) can lead to charge ringing back and Avoid high di/dt – move to softer (slower) switching forth, typically at very high frequencies. or zero-current switching where possible. Ringing between parallel rectifiers. Depending on Make the impedance as low as possible in the small variations of the turn-off voltage, the reverse signal processing nodes – implement current-based recovery charge (or parts of it) can flow back and forth signal transfer or add additional resistors to ground at between the two diodes, typically at high frequencies. sensitive inputs (if possible) to reduce the impedance of Noise pickup in secondary side chokes. Converters these nodes (easier when the signal is driven by a requiring a secondary side choke can pick up magnetic voltage source). Differential signaling should also be fields from the transformer and other primary side considered. The induced noise voltage should couple circuitry. This is best avoided by carefully arranging the into both signals identically and is then factored out at magnetic orientation of the components. the receiver, adding circuit complexity.

Fairchild Semiconductor Power Seminar 2010-2011 9

V. EMI AS AN INTEGRAL PART OF THE DESIGN FLOW

Design step Measure

Specification Define required EMI levels the power supply must comply with

Choose topology that creates low EMI (e.g. QR flyback for lower powers, LLC for higher Select topology powers)

Implement enough headroom for the parasitic ESR of the EMI filter. Calculate the components Make sure the nodes, especially those inside the control loop, have the lowest possible impedances.

Use a simulation model for the LISN to predict the EMI generated by the power supply, separately for common and differential mode noise. Simulate the design Choose the EMI filter topology based on the required attenuation levels. Calculate a first version and simulate this too.

Try to be close to the final arrangement of components in the finished power supply, so the radiated EMI signature can be verified. Build a prototype Minimize high-current loop areas and parasitic capacitance of nodes with high dv/dt. Leave some space at the input to put in an EMI filter later.

Once the power supply works well inside the specifications, perform pre-compliance testing of the power supply without an EMI filter, to measure the conducted noise (again, common and Test the prototype differential mode), and compare with the simulation results. Perform first radiated EMI test to identify potential "candidates" for radiated fields.

Build the EMI filter into the prototype and perform another full function and EMI pre- Add the EMI filter compliance test. Note that the EMI filter adds impedance at the power input and may resonate, so the regular function of the power supply must be verified again.

The implementation of the final version in a PCB will change certain aspects of the conducted Design the final and radiated EMI . version Make sure the specification for the components used in production comply with the requirements for being EMI compliant (particularly the bulk caps and inductors).

After verifying the full function of the power supply, perform pre-compliance testing to ensure that the noise levels are still acceptable. Perform pre-compliance testing at high and low line conditions, with low line being more Test the final critical (longer conduction angle on the input rectifiers). version Examine the impact of different component sources on the noise. Build several power supplies and check if the noise levels are repeatable, or vary significantly from power supply to power supply.

Fairchild Semiconductor Power Seminar 2010-2011 10

1. Write the Specification Hard-switched topologies have a difficult EMI At this point, the designer needs to determine the behavior as they generate high dv/dt and di/dt regulations governing the sale and use of the application. during switching. Since such topologies are used at In most cases, this will be EN550xx and EN61k, as higher power levels, the differential noise can be mentioned before. Here, it might be important to substantial, requiring large filters with low consider the project timing needed for compliance impedance at line frequency but high impedance at testing, and the space requirements for EMI filtering and noise frequencies. shielding. 3. Calculate the Components 2. Select the Topology When calculating the components of the power supply, Ongoing improvement in power switch technology two factors should be considered. First, the estimated and control IC functionality has led to a change in parasitic impedance of the EMI filter is part of the input topologies, especially recently. The winners emerging in impedance of the power supply. This is the case this topology race are the LLC converter for higher especially at low-line conditions, where the current is power levels, quasi-resonant flyback converters for highest but the available voltage lowest. Realistic values lower power levels, and primary-side regulated must be assumed to ensure a working prototype close to converters for very low power levels. An overview of the final design. topologies versus typical power levels is shown in Second, to improve sensitivity against noise for proper Figure 14. functioning of the power supply, consider making the impedance of nodes as low as possible, particularly inside the feedback or protection circuitry. As a result, higher noise power is required to disturb the circuit. Additionally, certain signals can be moved from voltage to the current domain, or differential sensing can be used, routing the two signal wires as closely together as possible. Here, a differential-mode receiver will be able to detect the wanted signal even in the presence of noise. Fig. 14. Selecting low EMI topologies. This method adds complexity, but can be a good solution Both the quasi-resonant flyback and LLC converters in very dense environments. have good EMI behavior, making them the topology of 4. Simulate the Design choice for many modern applications. Other topologies can be used but have some disadvantages. The regular Modern circuit simulators are fast and easy to use. flyback topology creates very high dv/dt and a high Since the noise signature of a power supply depends on voltage peak. Here EMI compliance is not easy due to many factors, including final layout, size and material of the high common mode noise components. In a flyback the components, housing, and surrounding equipment, converter, the transformer is also the energy-storing simulating the noise behavior is not very exact. However, component, resulting in high inductance values. The it can give a good indication of what may need to be most effective way to achieve this is to use a gapped done later. Figure 15 shows an example of a test fixture core. The magnetic fields emanating from the gapped used in a SPICE simulator: core are also a source of radiated EMI.

Fairchild Semiconductor Power Seminar 2010-2011 11

Fig. 15. Simulation schematic for EMI in switching regulator. A sine wave voltage source V1 models the line input As expected in a power supply without PFC, the input voltage. The LISN is represented through its schematic, current spikes at line frequency are large (3.5A). It is and the two measurement voltages are labeled VdiffL and obvious that they also contain high harmonic content at

VdiffN. At the center is a two-stage EMI filter in T- multiples of the line frequency, as shown in Figure 17. topology with a common mode choke and two Y-caps, followed by a bridge rectifier and bus capacitor, in parallel with a load resistor and a current source that will force a rectangular current into and out of the capacitor. The parasitic capacitors C7 and C10 simulate the coupling of switching noise into the ground line. Simulating the load is not very difficult. Here, a load current of +/- 0.5A has been assumed, with a period of 10µs (or 100 kHz switching frequency), and an on-time Fig. 17. Simulated EMI spectrum. of 4.9µs. The rise and fall times are assumed to be 100ns. Here, the frequency range of 50Hz to 2kHz is shown This is not far away from an LLC converter operating at as per the standard. Clearly, this design would not pass 150W output power. EN61k testing. The simulator would allow the insertion First, the simulation of the current at line frequency is of low-frequency PFC filters to reduce the amplitude of shown in Figure 16. the higher harmonics as needed but at this power level and the low frequencies involved, the filter would be very large. For this reason, an active PFC circuit should be considered.

Fig. 16. Simulated input current and voltages. Fig. 18. Simulated spectrum without EMI filter.

Fairchild Semiconductor Power Seminar 2010-2011 12

Figure 18 shows the spectrum of VdiffL when no EMI 5. Build a Prototype filter is inserted in the circuit. The main switching As explained, the noise signature of the power supply frequency at 100kHz is clearly visible, and higher depends on many factors, including the layout and harmonics and the intermodulation results show up at geometrical arrangement of the components. Therefore, higher frequencies. The amplitude is decreasing, which it is desirable to build the prototype close to the final is caused by the decreasing impedance of the bus outline of the power supply, while still leaving space for capacitance. In fact, playing with the ESR and ESL the test points and the EMI filter. Try to be close to the values of the bus cap show its impact on the amplitude final arrangement of components in the finished power of the noise peaks. From this picture, a first indication of supply so the radiated EMI "signature" can be verified. the required attenuation levels can be derived and Additional insight comes from measuring the different filter choices can be made. inductors being inserted, like the transformers and chokes. This is done with a network analyzer, checking the behavior up to say 30MHz. With inductors, the parasitic capacitance and the inductance value at high frequencies is interesting and can help to spot radiated EMI issues. With transformers, the leakage inductance (also at higher frequencies) and the parasitic

Fig. 19. Simulated spectrum with T-filter. capacitances are at the primary, from primary to secondary, and to the shield may be helpful to know. Fig. 19 shows the spectrum with the T-filter inserted. The routing on breadboard-type prototypes is usually The main switching frequency is still visible, but the done with hand-soldered metal wires that may have a higher frequency content has been greatly attenuated. lower resistance than the traces on the PCB of the final Simulating this design on a regular PC takes less than a implementation. This may appear to have little relevance, minute, so testing alternative filter circuits and but remember that the limits for noise voltages in component values is relatively easy. The main advantage question are very low as well. The higher PCB trace of using a behavioral representation of the main resistance leads to a higher noise voltage and trying to switching action of the power supply is the significant reduce it by widening the traces may result in higher reduction in simulation time. common-mode noise. It can help to damp the ringing It is important to add realistic parasitics to the that may occur. elements used in the EMI filter. If the simulation model In the prototype, it is important to minimize high- uses only ideal inductors and capacitors, the simulation current loop areas by putting the components close is pointless. In this example, the inductors have a together, and routing the wires as close to each other as parasitic C of 100pF and the capacitors have a parasitic possible. This area is the “transmit” coil of the radiated inductance of 20nH. This may sound large but given the field. If this area is minimized, then the field will be later arrangement of the components inside the filter, and minimized and the parasitic inductance in this loop will the PCB traces to connect the parts, the values are of the slow down switching unnecessarily, leading to lower right order of magnitude. Datasheets can provide a good efficiency. If possible, the wires should be routed on top indication of realistic parasitic elements. Alternatively, a of each other. It is best to start the layout considerations good network analyzer can calculate the parasitic with the high-current loops and arrange the rest around it. elements to be used in the models. The control IC and its associated feedback and control circuitry should be arranged such that magnetic fields influence them as little as possible.

Fairchild Semiconductor Power Seminar 2010-2011 13

The power supply will have circuit nodes with a high In addition, check for frequencies that are not dv/dt (the switching nodes). These nodes should be kept harmonics of the switching frequency – if you can spot as small as possible, since any parasitic capacitance here them in the pre-compliance EMI spectrum, they may be will couple the switching noise capacitively to ground. a problem later. These frequencies come from resonant To test this, mount the PCB on plastic spacers and place tanks in the circuit that are triggered by the switching it on a metal surface connected to the ground of the input action of the power supply, but resonate at their own connector. This setup should be similar to the power natural frequency. Typical candidates for this are: supply mounting in the final product. The common - Inductors resonating by themselves, at high mode noise will be much higher compared with testing frequencies where the parasitic capacitance and the the power supply on an isolated bench with a plastic inductor without core resonate (at these frequencies, surface, where the coupling will be much smaller. the core may no longer contribute) Finally, leave some space at the input of the prototype - Parasitic trace inductances (as a rough estimate, for placement of EMI components. In most cases, a 1nH/mm) and parasitic capacitances of switching common mode choke, two additional differential mode devices (e.g. MOSFET output capacitance or diode inductors and two capacitors across the lines should be a capacitance) good start. The common mode capacitors should be - Transformer leakage inductance and parasitic included as well. Figure 20 shows this basic setup: capacitances of the switching devices First, perform radiated EMI tests to identify potential "candidates" for radiated fields. This procedure is explained in the section on radiated EMI. It is worthwhile spending half an hour on the working power supply, running at full power if possible (since this will cause the highest current spikes) and checking if particular hotspots can be found and explained. Next, Fig. 20. Prototype EMI filter for 30W -200W Power Supplies compare the detected EMI frequencies with the lines in This should work for power supplies in the range of the conducted EMI spectrum to verify if they show up 30W to 200W. For smaller power supplies, fewer there as well. It is possible to get a “clean” voltage components may be required. For larger power supplies, waveform on a node that shows strong radiated EMI, so a larger filter is needed, mostly driven by the higher peak looking at scope shots may not reveal the full picture. currents. Remember that when the core of the inductor Redo the test at lower power, especially when the power goes into saturation, the effective inductance value is supply changes its working modes across the power greatly reduced, and all the noise will pass. range, which many do (e.g. disabling parts of the power supply at lower power). 6. Test the Prototype Once the power supply works well inside the 7. Add the EMI filter specifications, it is worth performing pre-compliance Now it is time to add the pre-calculated EMI filter and testing of the power supply without an EMI filter, test again. First, test the power supply across the full measuring the conducted noise (again, common and load range to see if the EMI filter has any negative differential mode) and checking against simulation impact on the proper function as it adds impedance in results. This will give a first indication of the required the input. If this is not done well, the added impedance filter attenuation. It is quite useful to perform this test at line frequency can starve the SMPS, and the output without an EMI filter, as this will show the unfiltered impedance of the EMI filter and its resonances may behavior. By adding the standard EMI structure, the cause irregular behavior. potential problems may be harder to spot, since they may Next is another pre-compliance test on conducted EMI be smaller in amplitude. to see if the filter actually works as designed, and

Fairchild Semiconductor Power Seminar 2010-2011 14 provides enough attenuation, with a margin of 6dB since they tend to have less variation. It may be 10dB. If not, check the spectrum and modify the advisable for correct high frequency behavior to put a component values. Another effective method, as foil capacitor in parallel with the bus capacitor, since this explained above, is to add series-resonant tanks or RC will control its high frequency behavior much better. shunts to help at certain problem frequencies. This is also the topology you usually end up with when If the damping is too high, it may be advisable to combining an EMI filter in a pi topology with a reduce the component values, e.g. by removing turns on “normal” SMPS. the inductors, since the higher impedance may Inductors may also vary significantly. Here, it is unnecessarily restrict dynamic behavior. mostly the mechanical changes like proper mounting of Remember to perform the test in a mechanical the two halves of the transformer core that can make a environment close to the final product, particularly for big difference. common mode noise, which is predominantly driven by capacitive coupling. This can be done with a metal sheet 9. Test the Final Version just below the PCB (with some isolation, of course). On the final version of the power supply, after verifying the full function of the power supply, perform 8. Design the Final Version pre-compliance testing again to check if the noise levels The implementation of the final version into a PCB are still acceptable. You should have a good will change certain aspects of both the conducted and understanding on the areas of influence you still have radiated EMI so care is needed when applying the when something is not as it should be. modifications. Specifically, the grounding and the Pre-compliance testing at high and low line conditions mechanical arrangement are important. The grounding should be performed, with low line being more critical will impact the common mode noise and the mechanical (longer conduction angle on the input rectifiers and arrangement will impact radiated EMI and its impact higher currents in the power supply). This will be the within the power supply. Try arranging larger chokes at worst case for radiated EMI. 90° angles to reduce the magnetic pickup of the noise. In Build several power supplies and check if the noise a flyback converter, the transformer shield should be levels are repeatable, or vary significantly from power connected to the source of the switching element to supply to power supply. If so, the real source for the shorten the capacitive return path of the switching noise. noise signal(s) has not properly been identified and If a second shield is implemented, it should be connected controlled. As most components cannot vary that much to ground. Watch out for changes in the pinout of during operation, a prime candidate for this particular connectors or high-current components, as this may problem is the gate drive of the main switches. If there seem like a little change but can modify the high-current are layout issues, ringing, spikes, or capacitive feed- loops and their radiating area. through, the rise and fall times may not be well- It is also time to think about manufacturing and the controlled, causing larger variations in the EMI spectrum. specifications of the components to be used in mass Again, you may not be able to see this from the voltage production. Their characteristics may vary (especially waveforms on the switching nodes, and this problem with the passive components), and so will their behavior may also not be apparent from the efficiency curves. regarding EMI. It may also be advisable, especially when using It is quite common for capacitance values of passive components that are not fully specified and electrolytic caps to change between -20% and +50%. guaranteed in performance, to test a number of Few manufacturers will specify variations on ESR and components of the different sources, to see if the power ESL, yet these values have a strong impact on EMI supply fully works and complies with EMI requirements behavior, as seen before regarding the bulk capacitor and at any combination. conducted EMI. Component variations are less of an issue with a (foil) capacitor used e.g. in the EMI filters,

Fairchild Semiconductor Power Seminar 2010-2011 15

VI. CONCLUSION REFERENCES As long as voltages and currents are being switched, [1] Bob Mammano, Bruce Carsten: Understanding and optimizing EMI will be generated and need to be addressed. This electromagnetic compatibility in switch-mode power supplies (TI power seminar series 2003) implies that there will never be a “silver bullet” — just [2] Bruce Carsten: Application note for H-field probe (http://bcarsten.com) improvements to the situation to arrive at an acceptable [3] Jonathan Harper: Electromagnetic compatibility design for power supplies (Fairchild Semiconductor power seminar series 2004/2005) compromise. Once the basic mechanisms are understood, [4] Richard Lee Ozenbaugh: EMI filter design (CRC, Nov 2000) [5] Christophe Basso: Switch-Mode Power Supplies SPICE Simulations and it is easier to analyze and re-engineer a given power Practical Designs“, McGraw-Hill, 2008 supply to improve its behavior and really exploit all the Alfred Hesener is a Director with Fairchild Semiconductor, performance advantages of modern power switches. responsible for Applications and Marketing in Europe, covering all applications and products, with a focus on power electronics in industrial and automotive applications. Previous ACKNOWLEDGEMENT experience includes analog design and technical marketing for power management with Infineon Technologies, and field The author would like to thank Dr. Michael application engineering with Atmel. Totaling 16 years of Weirich for his sharing his experience in many experience, he holds an MS-EE in semiconductor electronics and three patents in power semiconductors. Main area of interest is in ecodesign and system discussions, his very valuable insight and support, design considerations. in particular on how to make theory match reality.

Fairchild Semiconductor Power Seminar 2010-2011 16